Agent-native video toolchain. Local-first video editing for agents and humans. CLI and MCP, browser review, plain files on disk.
Agent Claude Code
Docs-to-code-truth sync on Sonnet. Use after features land to update README, AGENTS.md, CLAUDE.md, and CHANGELOG from verified facts. Measures counts itself and files honest gaps as Linear issues (never a repo TODO).
Agent Claude Code
Red-green TDD implementer on Sonnet. Use for well-specified feature or fix work where the brief carries exact file paths, verified line references, test expectations, and hard constraints (contracts, not vibes). Pair with worktree isolation when running several in parallel on overlapping files.
Agent Claude Code
Adversarial cross-feature reviewer on the session model (highest judgment tier, one per ship). Run AFTER the specialist reviewers, feeding it their findings so it hunts what they MISSED - cross-cutting races, lifecycle and process-boundary failures, schema evolution hazards, and interactions BETWEEN features.
Agent Claude Code
Specialist code reviewer on Sonnet. Dispatch one per review dimension (testing, security, performance, api-contract, maintainability, design, or a general checklist pass) against a diff, several in parallel. Emits machine-mergeable JSON-line findings under a strict evidence gate.
Agent Claude Code
Read-only reconnaissance on Haiku (cheap, fast). Use PROACTIVELY before any implementation wave to map integration points, gather file:line facts, and inventory conventions, APIs, and test patterns. Dispatch several in parallel with narrow briefs. Never for judgment calls, design decisions, or code changes.